Centre for Digital Humanities

Events

HAI Colloquium with Prof. Rens van de Schoot

Event details

Date:
28 October 2021
Time:
13:15 - 14:15
Venue:
Drift 21
Drift 21, room 0.32, Utrecht

The focus area Human-Centered AI (HAI) organizes the first Human Centered AI colloquium of this academic year with Prof. Rens van de Schoot as speaker.

Prof. van de Schoot works as full professor ‘Statistics for Small Data Sets’ at Utrecht University and is extra-ordinary professor at the North-West University in South-Africa. He is also program director of the research master ‘Methodology and Statistics for the Behavioural, Biomedical and Social Sciences’ and coordinator of the post-graduate program at the department of Methods and Statistics at Utrecht University.

He will talk about: Saving time and sanity – Active learning for screening large amounts of texts.

The scientific output of the world doubles every nine years. Imagine updating a medical guideline, making evidence-based policy or scouting for new technologies in this tsunami of new knowledge: there’s not enough time to read everything. More and more researchers and organizations rely upon Systematic Reviews to synthesize the state of the art in a particular scientific field. But this process is under severe strain: the exponential growth in papers means they too have to screen an ever-larger body of work – resulting in costly, abandoned, or error-prone work. A well-established approach to increase the efficiency of title and abstract screening is screening prioritization via active learning. Active learning is effective for systematic reviewing, and we can reduce the number of papers to screen up to 95%(!). Over the last years, we have worked with a multidisciplinary team to develop and validate an Open Source tool to support literature research in quickly finding relevant articles: ASReview – with >150K downloads of the user-friendly front-end in >40 countries, and is published in Nature Machine Intelligence. The software is designed to save time and improve transparency when screening large amounts of text using state-of-the-art machine learning models and feature extraction techniques. The tool is easily expandable, and many master students have already added new models to the software.

In the colloquium, I will introduce active learning for screening textual data using a real use-case of developing medical guidelines to deal with the Covid19 pandemic. Then, I will explain the difference between Human-in-the-loop and Research-in-the-loop approaches, introduce the underlying machinery, present results from simulation studies, and show how former master students have contributed to the open-source project. Lastly, I will explain the upcoming hackathon we are organizing for Follow the Money, investigating the communication between the multinational oil and gas company Shell and the Dutch government. They have to screen thousands and thousands of emails, let’s see if we can help them to speed up this process by implementing active learning!